Wednesday, May 1, 2002

Fixed SVG build bustage for the reflow tree.

Thursday, May 2, 2002

God I'm sloppy. I should've run the reflow tree build with all my shiny new assertions before checking that in. It turns out that we assert-botch in a benign way in several places, and I've now fixed those. rbs also pointed out some problems with MathML incremental reflow: he referred me to the MathML editor, and I had to do a bit of polish on the nsMathMLContainerFrame. Cut a new branch, REFLOW_20020502_BRANCH containing the latest and greatest changes.

Spent some time looking at (and even may have fixed!) a floater incremental reflow bug that occurs because of bad interaction between the block-and-line code and the table code. Looks like it might also fix the other problem I was looking at a few days ago, as well.

Monday, May 6, 2002

• Filed a bug for doing {ib} during reflow: I'm secretly hoping that kin will take it.
• Proposed that we back out a misfeature that causes layout to re-enter itself.
• Looked into a template builder topcrash at dbaron's prompting. It looks like something is going wrong in nsXULContentBuilder::CreateTemplateContents: either we're walking off the end of the content model without finding the resource element, or we don't have the resource element in the content map.
• Spent some time looking into a floater bug that's visible on amazon.com. We're getting a `bad floater placement' assertion, which makes me think that it's the same thing as bug 76135. Anyway, it turns out that there's a case where we're not removing floaters from the space manager before flowing a block frame for a second time. I proposed a fix, but I think it will need some work.

Wendesday, May 7, 2002

• Long code review with kin for the reflow tree.
• bzbarsky found a problem with the reflow tree and a <select> test case.

Thursday, May 9, 2002

I've fixed the incremental reflow problem with <select> frames that bz pointed out (breaking the test case in bug 82873). Looks like we'll need that code I ignored in nsComboboxControlFrame, after all!

kin noticed some assertions firing with the sidebar. Caught a case where the nsHTMLFrameOuterFrame needed to convert an incremental reflow to a dirty if the child wasn't on the reflow path. Also, needed to tweak the nsBoxToBlockAdaptor to pass back the `fixed' reflow path from CanSetMaxElementSize since RefreshSizeCache can cause the adaptor to reflow its frame.

kin also discovered that resizing a frame group would cause a crash. I think I checked the nsHTMLReflowState ctors in every subdirectory of layout except for layout/document. Oops.

After thinking about it some more, I think that I'm a dolt for ignoring rbs's suggestion. I've changed the logic in the nsHTMLReflowState ctors so that if the child frame isn't along the path, it automatically converts the reflow reason from `incremental' to `dirty'. This allowed me to undo the reflow reason twiddling in MathML, as well as some other places. It also makes it a lot harder to walk off the end of the reflow path by mistake, which should mean I'll have fewer crash bugs to deal with when this gets checked in.

Tuesday, May 21, 2002

Karnaze found a regression from the reflow tree changes.

Friday, May 24, 2002

Added some bullet-proofing to the RDF container and container-utils stuff to deal with a topcrash in the NS7 beta. Fixed a crasher dealing with fieldset frames.

Tuesday, May 28, 2002

Gave a whirl at fixing a 4.x quirk with <P> elements inside lists. Unfortunately, it looks like it's a symptom of a deeper problem. Tried to fix another 4.x quirk with content inside <th> elements; maybe it'll stick.